B3LYP simulations for CNTs

[manas_pratap]

Hi all,

I've neem trying to obtain a bandstructure using B3LYP for a few CNTs, but I have been unsuccessful so far.
I first do a standard DFT SC calculation to obtain the CHGCAR and WAVECAR files. I then copy these files into the B3LYP folder, where I create a KPOINTS file suitable for HF calculations (weighted and zero-weighted kpoints)
I then use the following INCAR file for the calculation. However, my bandstructure does not look like a bandstructure at all.

I would be grateful for any advice on this matter. Thank you!

Code: Select all

ISMEAR =  0
SIGMA  =  0.05
LHFCALC = T
GGA = B3
AEXX = 0.2
AGGAX = 0.72
AGGAC = 0.81
ALDAC = 0.19
ALGO = Damped
TIME = 0.4
LREAL = AUTO
EDIFF = 1.0E-07
ENCUT = 600.00
KPAR = 3
NCORE = 8
NBANDS = 256

I've considered NBANDS = 256 since I have 32 atoms in my system, all carbon atoms.

POSCAR -

Code: Select all

CNT_8_0
    1.0
            21.29400000 0.00000000 0.00000000 
            0.00000000 21.29400000 0.00000000 
            0.00000000 0.00000000 4.26000000 
  C
  32
Direct
        0.35222061 0.50000000 0.16666667      	#	1      	C
        0.35222061 0.50000000 0.83333333      	#	2      	C
        0.64777939 0.50000000 0.83333333      	#	3      	C
        0.64777939 0.50000000 0.16666667      	#	4      	C
        0.50000000 0.64777939 0.83333333      	#	5      	C
        0.50000000 0.35222061 0.16666667      	#	6      	C
        0.50000000 0.64777939 0.16666667      	#	7      	C
        0.50000000 0.35222061 0.83333333      	#	8      	C
        0.55655273 0.63653036 0.66666667      	#	9      	C2
        0.55655273 0.36346964 0.33333333      	#	10     	C2
        0.44344727 0.63653036 0.33333333      	#	11     	C2
        0.44344727 0.36346964 0.66666667      	#	12     	C2
        0.36346964 0.44344727 0.33333333      	#	13     	C2
        0.36346964 0.55655273 0.66666667      	#	14     	C2
        0.63653036 0.44344727 0.66666667      	#	15     	C2
        0.63653036 0.55655273 0.33333333      	#	16     	C2
        0.44344727 0.36346964 0.33333333      	#	17     	C2
        0.44344727 0.63653036 0.66666667      	#	18     	C2
        0.55655273 0.36346964 0.66666667      	#	19     	C2
        0.55655273 0.63653036 0.33333333      	#	20     	C2
        0.63653036 0.55655273 0.66666667      	#	21     	C2
        0.63653036 0.44344727 0.33333333      	#	22     	C2
        0.36346964 0.55655273 0.33333333      	#	23     	C2
        0.36346964 0.44344727 0.66666667      	#	24     	C2
        0.60449581 0.60449581 0.83333333      	#	25     	C3
        0.60449581 0.39550419 0.16666667      	#	26     	C3
        0.39550419 0.60449581 0.16666667      	#	27     	C3
        0.39550419 0.39550419 0.83333333      	#	28     	C3
        0.39550419 0.39550419 0.16666667      	#	29     	C3
        0.39550419 0.60449581 0.83333333      	#	30     	C3
        0.60449581 0.39550419 0.83333333      	#	31     	C3
        0.60449581 0.60449581 0.16666667      	#	32     	C3

[manuel_engel1]

Hi,

Thanks for reaching out. I will try to assist you as best as I can. To do that, I would like to ask a few clarifying questions first:

• When you say that you want to obtain a band structure, do you mean a band-structure plot along high-symmetry lines?

• What exactly does not look right about your band structure? What are your expectations?

• Do you obtain a good band structure when you do regular DFT instead of hybrids/HF?

• Could you please upload a minimum working example including the relevant input and output files according to the forum posting guidelines? I would like to inspect your calculation and be able to reproduce it if necessary.

[manas_pratap]

Hello, thank you for responding.

1. Yes, I mean a band-structure plot along high-symmetry lines. However, since transport in CNTs are along one direction only, I would need the high-symmetry lines in that particular direction.

2. The valence bands are looping. Here's an example of what's wrong with my bandstructure -

3. I get a reasonable bandstructure when I do standard DFT, as well as with HSE06. My problem in particular lies with the B3LYP calculation.

Here's a minimum working example. I will also include the input files for the B3LYP calculation, but these take a very long time to finish.
I am grateful for any advice you may have.

Thank you.

[manuel_engel1]

Thank you for providing the additional information. Unfortunately, I'm still not sure what is causing the issue you are facing. Could you please also share the OUTCAR files of the individual runs, in particular the band-structure calculation with B3LYP?

Do you copy the output of the B3LYP calculation to the bandstructure directory and rerun non-selfconsistently, the same way you do it for standard DFT?

I am not an expert when it comes to CNTs but I don't see a reason why B3LYP wouldn't work for this system.

[manas_pratap]

Hello,
Sorry it seems I forgot to include the OUTCAR files.
I'm attaching the files here.

For B3LYP calculations, I do the same procedure as with HSE06, where I copy the output from the self consistent run into the bandstructure folder.

[manuel_engel1]

No worries, thank you for uploading the file.

I have quickly looked at the OUTCAR file included in the Standard_bandstructure directory (which seems to be the B3LYP band-structure calculation along a k-path). When plotting the eigenvalues directly from the OUTCAR, I get a reasonable band-structure plot that looks nothing like the image you have posted earlier. This seems to suggest that there is nothing wrong with the calculation but with the processing of the output. I think we are getting much closer to the answer. Here are some things to consider:

• How do you extract/plot the band structure (scripts, py4vasp, etc.)?

• From which output file are you reading the band structure (OUTCAR, EIGENVAL, PROCAR, vasprun.xml, vaspout.hdf5, etc.)?

• Are you able to confirm that the energies as listed on the OUTCAR or EIGENVAL files are OK? Or do they show the same strange behavior?

In the case that you are unfamiliar with the OUTCAR file, the relevant section starts like this:

Code: Select all

 E-fermi :  -2.9683     XC(G=0):  -3.4468     alpha+bet : -2.8602

 val. band max:       -2.9803326213   @ k =   0.0000   0.0000   0.1050
 cond. band min:      -2.9614155711   @ k =   0.0000   0.0000   0.2101
 fundamental gap:      0.0189170503
 Fermi energy:        -2.9682526466

 k-point     1 :       0.0000    0.0000    0.0000
  band No.  band energies     occupation
      1     -22.1627      2.00000
      2     -20.3023      2.00000
      3     -20.3023      2.00000
      4     -16.7842      2.00000

and it continues to list all k-points and bands.

[manas_pratap]

Hello,

Thank you for your response.
The standard_bandstructure directory contains the output data for the bandstructure calculation that does not use B3LYP. The output files for the B3LYP bandstructure calculation are in the B3LYP_Bandstructure directory.

For plotting, I use a script that reads the data from the vasprun.xml file. The standard_bandstructure plot actually looks fine, which you've confirmed as well. The issue is with the B3LYP bandstructure.

Edit: I went through the OUTCAR file in the B3LYP_bandstructure directory. There are some k-points with very large eigenvalues (negative). I think this is causing the problem.

[manuel_engel1]

It seems to me that the calculation inside B3LYP_bandstructure is a self-consistent calculation using an automatically generated k-point mesh. If I understood everything correctly, that means that you still need to run the actual band-structure calculation. Just as with the basic DFT calculation, you have to read the converged charge density and wave functions into a new VASP calculation by setting ICHARG=11 and providing a k-points file in line-mode. You have done this successfully already for the basic DFT calculation. It works the same way for the B3LYP functional.

Let me know if that solves your problem.

[manas_pratap]

Hello,
Thank you for your response. The B3LYP_Bandstructure calculation follows the Standard DFT+HF calculation outlined here -
wiki/index.php/Bandgap_of_Si_using_diff ... HF_methods

If I understand the procedure correctly -
1. A standard PBE calculation is done first. This yields converged CHGCAR, WAVECAR files and an IBZKPT file.
2. The CHGCAR file and WAVECAR files are copied into the B3LYP calculation directory, and then the KPOINTS file needs to contain weighted and zero-weighted k-points. The zero-weighted kpoints can be generated by tools such as https://www.materialscloud.org/work/tools/seekpath, and for the weighted k-points, I would copy the IBZKPT file contents.
3. I then run the B3LYP calculation, and plot the bandstructure. This is according to the VASP Wiki.

I'm not sure how I can do a B3LYP charge density calculation without the KPOINTS file that contains both weighted and zero-weighted kpoints file.

[manuel_engel1]

Hello again,

You are correct of course. I misinterpreted your input/output files. Unfortunately, I cannot yet tell where exactly the problem lies. I will have to make some runs of my own using your input files as templates. I will get back to you as soon as I have any new information. Thank you for your patience.

[manuel_engel1]

I have tried to reproduce this issue by running a smaller system with fewer k-points. In this case, I was not able to reproduce your findings. I have also started the B3LYP calculation that you have supplied but it is taking some time until it is finished. Do you see the same behavior also for other systems/functionals? Is it possible to reproduce this error in a smaller system that I could run quickly on my end?

[manas_pratap]

Thank you for your post. I appreciate all the help.
Here's the zipped folder containing all B3LYP material for a smaller structure.

[manuel_engel1]

This is quite a tricky issue to figure out. I have now run your smaller example with different settings. I was able to reproduce the issue using your exact input. However, after switching from B3LYP to HSE and using the same input parameters as in the Si bandstructure tutorial, the band structure looks OK.

It takes some time to run these calculations, so I wasn't able to test a lot but I still wanted to return to you with a partial answer. I'm not yet sure what exactly is causing the weird behavior, but I will try to do a bit more investigating.

[manas_pratap]

Hello,

Thank you very much for your response. I had also run HSE calculations for these systems and they turned out to be fine.
Thank you for your efforts to solve this problem. In the meantime, I've tried tweaking a few parameters to see if it makes a positive difference to the B3LYP result, but unfortunately I still get the same problem.

[manuel_engel1]

After talking to the team, there are two things that I would like to recommend:

• Set HFRCUT=-1 in the B3LYP band-structure calculation. The default may not work for calculations that are not on a strictly regular k-point mesh.

• Use the KPOINTS_OPT file to calculate the band structure along a path, as described on this wiki page.

Let me know if any of these things work for you.